Bug Report: Incorrect Title Extraction for Documents with Front Pages

[yashp048]

Description

ScienceBeam fails to correctly extract article titles from PDF manuscripts that have front pages (cover pages with journal branding, terms & conditions, etc.). Instead of extracting the actual article title, it extracts header/footer text and journal information from the front page.

What are the Steps to Reproduce to issue?

1. Submit a manuscript PDF with a front page containing journal branding and terms/conditions (e.g., Taylor & Francis "Expert Review" journal format)
2. Process the document through ScienceBeam for XML conversion
3. Examine the resulting XML file's element in the // section

What is the Expected behaviour?

The element should contain the actual manuscript title. For example, based on the content in file 76_pdf-0.1.11.xml(See attached file)

76.pdf, the title should be something like: "The Toxic Effects of Ethylene Glycol Tetraacetate Acid, Ferrum Lek and Methanol on the Glutathione System: correction Options"
But it is coming as(see screenshot) -

Additional Context

Affected file:

76_pdf-normal.xml

The actual article content (abstract, body) is correctly extracted
This appears to be a pattern recognition issue where ScienceBeam is not properly identifying and skipping front page content when locating the title

[de-code]

Thank you for reporting this.

The current default models have all been trained on bioRxiv/medRxiv preprints only which would explain it not handling those kind of papers (although the title is also on the first page).

ScienceBeam Parser also supports loading GROBID models that are trained on more varied papers, but that also didn't seem to extract the title correctly. I tried this HF Space using GROBID 0.8.0 models instead:
https://huggingface.co/spaces/de-code/sciencebeam-parser-grobid-0-8-x-wapiti-models

[yashp048]

@de-code
I have also seen a pattern that whenever front page has footers, the parser extracts the footer text and make it prefix of the title.
Any idea why?